An unrooted phylogenetic network of the 315 rice samples is presented in . As could be expected, Oryza sativa ssp. indica, O. sativa ssp. japonica and O. glaberrima form three distinct clusters. Nerica varieties of interspecific origin align along the japonica axis, with Nerica 1 and 2 facing the O. glaberrima branch. In addition to these three clusters, a fourth distinct cluster, consisting of two sub-clusters, was observed, at the junction of the O. glaberrima-indica-japonica axes.
Phylogenetic relationships among the 315 samples studied.
Analyses with the software ‘Structure’ showed that the major structure in the data was captured when four populations were assumed. Three of these populations corresponded with Oryza sativa ssp. indica, O. sativa ssp. japonica and O. glaberrima, respectively, while the fourth population corresponded with cluster 4 in . Of the 315 materials 285 samples were assigned to a cluster with more than 91% probability. All materials in cluster 4 in were assigned to cluster 4 with more than 81% probability in Structure, except two varieties from Senegal that were assigned to cluster 4 with 59% and 46% probability.
Prior to the molecular analysis, all varieties collected from farmers were classified as O. sativa, O. glaberrima, hybrid or unclear. None of the materials assigned to the two O. sativa clusters with more than 81% probability were classified as O. glaberrima and vice versa (). The single sample classified as O. sativa that was assigned to O. glaberrima, and the single sample classified as O. glaberrima that was assigned to O. sativa, were most likely caused by interchanging of materials during the experiment.
Presumed taxonomic origin of the 289 farmer varieties in relation to the assignment probabilities to the four observed clusters.
Cluster 4 comprised two subclusters (). All varieties in sub-cluster 4-2 had been taxonomically determined as O. sativa prior to the molecular study, while cluster 4-1 consisted of samples that had been determined either as O. sativa, O. glaberrima, hybrid or unclear (). The main distinctive features between these two sub-clusters were panicle stature at maturity and pericarp (or seed) colour. Sub-cluster 4-1 consisted of varieties with an erect panicle, typical for O. glaberrima (), or a semi-erect or slightly drooping panicle, and a red pericarp, except for a single variety from Senegal which had a brown pericarp. Farmers classify particularly the varieties with an erect panicle as O. glaberrima, because of the similarity in panicle stature. Farmers do not recognise the varieties of cluster 4 as a separate group. They divide all varieties into two types: those that resemble O. sativa and those that resemble O. glaberrima. Farmers are not specifically interested in varieties of interspecific origin, but in varieties that perform best under their conditions.
Presumed taxonomic origin of the farmer hybrid varieties observed in sub-clusters 4-1 and 4-2 in .
Main panicle types found in this study.
The three varieties in sub-cluster 4-1 that were classified as O. sativa
had semi-droopy panicles which made them less distinctive from O. sativa
. Sub-cluster 4-2 consisted of varieties in which panicles were predominantly strongly drooping, similar to O. sativa
, and in which the pericarp colour varied from white to brown (90% of the varieties had a brown pericarp colour). Except for pericarp colour, the varieties in sub-cluster 4-2 did not have any clearly distinctive morphological features from O. sativa
varieties (). Detailed morphological analysis of some varieties belonging to sub-cluster 4-2 in 2002 showed that when characteristics were aggregated in a Principal Component Analysis these farmer varieties were different from O. sativa
and O. sativa
ssp. japonica 
Main distinctive morphological features of 12 varieties from cluster 4*.
Genetic diversity within groups (He
) was calculated for each of the four clusters. For this purpose an assignment probability of 91% was used as cut-off point to define the four clusters. The He
value for cluster 4 was highest (0.098; n
40) followed closely by the He
value for the O. sativa
group (0.089; n
92). Relatively low values were observed for the O. sativa
group (0.045; n
87) and the O. glaberrima
group (0.034, n
Varieties in sub-cluster 4-1 not only displayed characteristics typical of O. glaberrima, such as the easily observable erect panicle stature (), but also characteristics of O. sativa, such as the long, pointed ligule typical of O. sativa (), a less conspicuous feature. The only explanation for this new morphotype is interspecific hybridization between O. sativa and O. glaberrima. This was supported by the molecular data, separating cluster 4 from O. sativa ssp. and O. glaberrima, and showing large within-group diversity.
Main ligule shapes found in this study.
Cluster 4 consisted of a considerable number of different farmer interspecific hybrids originating from the Upper West African coastal rice belt (). None of the modern varieties and none of the samples collected in Ghana and Togo were found in cluster 4 in , nor were any of these samples assigned to cluster 4 in with more than 40% probability. Thirty samples - originating from almost all countries, and including two modern varieties - were assigned with less than 91% probability to one cluster. No samples from Togo were assigned with less than 91% probability to one cluster. Although no samples from Ghana were assigned to cluster 4, five samples were assigned with high probabilities to two clusters. These samples may require further study to know whether they have an interspecific background. But we cannot assume that all such materials have an interspecific nature since one variety from IRRI was assigned to the O. sativa
group with 76% probability (Table S1
). Likewise, existence of samples with a very high assignment percentage probability does not rule out an interspecific origin. For example, WAB 450-I-B-P-105-HB, a Nerica that was never officially released was assigned with 100% probability to the O. sativa
Number of farmer varieties, modern varieties and (semi-) wild relatives assigned by the software ‘Structure’ to the four observed clusters.
To a certain extent, the sub-clusters relate to the countries of collection and local seed colour preferences. The varieties in sub-cluster 4-1 originate from Guinea Bissau (4), Guinea Conakry (2), Senegal (1) and Sierra Leone (14), while the varieties in sub-cluster 4-2 are from The Gambia (9), Guinea Bissau (6) and Senegal (9). Whereas in Guinea Conakry and Sierra Leone farmers commonly cultivate red rice (both African and Asian rice), farmers in The Gambia, Senegal and northern Guinea Bissau predominantly cultivate white rice. Southern Guinea Bissau occupies an intermediate position, as red rice is still cultivated but farmers strongly prefer white rice.